Pytorch Maml

PyTorch implementation of MAML:
Alternatives To Pytorch Maml
Project NameStarsDownloadsRepos Using ThisPackages Using ThisMost Recent CommitTotal ReleasesLatest ReleaseOpen IssuesLicenseLanguage
Awesome Pytorch List14,103
3 months ago4
A comprehensive list of pytorch related content on github,such as different models,implementations,helper libraries,tutorials etc.
Nlp Tutorial12,403
5 months ago34mitJupyter Notebook
Natural Language Processing Tutorial for Deep Learning Researchers
External Attention Pytorch9,522
5 days ago2September 27, 202260mitPython
🍀 Pytorch implementation of various Attention Mechanisms, MLP, Re-parameter, Convolution, which is helpful to further understand papers.⭐⭐⭐
6 months ago178otherPython
Code for the paper "Jukebox: A Generative Model for Music"
8 months ago7mpl-2.0Jupyter Notebook
:robot: :speech_balloon: Deep learning for Text to Speech (Discussion forum:
Transformers Tutorials5,620
2 days ago209mitJupyter Notebook
This repository contains demos I made with the Transformers library by HuggingFace.
Awesome Domain Adaptation4,554
14 days ago2mit
A collection of AWESOME things about domian adaptation
a month ago88apache-2.0Python
The GitHub repository for the paper "Informer" accepted by AAAI 2021.
a year ago1apache-2.0Python
A collection of loss functions for medical image segmentation
3 years ago31Python
[CVPR2020] Adversarial Latent Autoencoders
Alternatives To Pytorch Maml
Select To Compare

Alternative Project Comparisons


This is a PyTorch implementation of the supervised learning experiments from the paper Model-Agnostic Meta-Learning (MAML):

Important: You will need the latest version of PyTorch, v.0.2.0 to run this code (otherwise you will get errors about double backwards not being supported).

Currently, only the Omniglot experiments have been replicated here. The hyper-parameters are the same as those used in the original Tensorflow implementation, except that only 1 random seed is used here.

5-way 1-shot training, best performance 98.9%

Alt text

20-way 1-shot training, best performance 92%

Alt text

Note: the 20-way performance is slightly lower than that reported in the paper (they report 95.8%). If you can see why this might be, please let me know. Also in this experiment, we can see evidence of overfitting to the meta-training set.

The 5-way results are achieved by simply meta-testing the network trained on the 1-shot task on the 5-shot task (e.g. for the 5-way 5-shot result, test the 5-way 1-shot trained network with 5-shots). Again the 20-way result is lower here than reported in the paper.

This repo also contains code for running maml experiments on permuted MNIST (tasks are created by shuffling the labels). This is a nice sanity check task.


This software is distributed under the MIT license.


  • [ ] port to pytorch 0.4 from 0.2 and python 3 from 2
  • [ ] investigate performance difference from TF version
  • [ ] add first-order version
Popular Pytorch Projects
Popular Paper Projects
Popular Machine Learning Categories

Get A Weekly Email With Trending Projects For These Categories
No Spam. Unsubscribe easily at any time.
Jupyter Notebook